Automatic discovery of concepts and actions
作者：
Highlights：
• Our method ADC simultaneously discovers relational concepts and learns behaviors.
• ADC automatically constructs its representation of the environment by exploration.
• ADC introduces novel mechanisms to automatically drive its own learning process.
• Knowledge incrementally learned by ADC in a task can be re-used in other tasks.
• Knowledge representation used in ADC is easy to understand by humans.
摘要
•Our method ADC simultaneously discovers relational concepts and learns behaviors.•ADC automatically constructs its representation of the environment by exploration.•ADC introduces novel mechanisms to automatically drive its own learning process.•Knowledge incrementally learned by ADC in a task can be re-used in other tasks.•Knowledge representation used in ADC is easy to understand by humans.
论文关键词：Relational representation,Intrinsic motivation,Reinforcement learning,Concept formation,Predicate invention,Discovery system
论文评审过程：Received 29 December 2016, Revised 23 May 2017, Accepted 9 September 2017, Available online 20 September 2017, Version of Record 13 October 2017.
论文官网地址：https://doi.org/10.1016/j.eswa.2017.09.023
